etcd: Scalability test methodology and tools

[ma...@mikitka.net]

Hello,

I need to deploy an etcd cluster with 200+ nodes, and I want to focus on the network effects of an application in this configuration on the smallest hardware profile possible. Is it possible to 'stub out'/'black hole' all write operations and respond to all read operations with a static set of values so that I have a host that is CPU bound? I would then deploy as many etcd nodes as possible on the same machine.

If this is not possible, I'll pursue lightweight containers.

My last resort is reserving 200+ nodes from a cloud resource provider.

Thanks in advance.